Subvolts take AI progress hinges on data efficiency, not just scale. Jun 19, 2026 - Developers
- Learn about the fundamental data challenges impacting AI model training and future development.
- Business
- Understand the core resource needs for AI progress, informing strategic decisions about data and model investment.
This is a solid interview exploring fundamental questions about AI development and data needs.
The data black hole at the center of AI
Audio streams from Dwarkesh Patel. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
Dwarkesh Patel hosts a 12-minute podcast episode titled "The data black hole at the center of AI." The discussion explores what truly drives AI progress, comparing human versus AI sample efficiency. It questions the significance of sample efficiency in the broader context of AI development.
This episode asks what really makes AI get smarter. It compares how quickly humans learn versus how fast AI learns from data. Understanding this helps explain why AI needs so much information to improve.
This episode probes the fundamental drivers of AI progress, a critical question for anyone following frontier labs like OpenAI, Google Gemini, and Anthropic. It's for researchers, developers, and AI enthusiasts interested in the core mechanics of model training. Understanding sample efficiency helps contextualize the massive data requirements of current large language models and future AI systems. The discussion offers a deeper perspective beyond just scaling up models, touching on the underlying data challenges that all major AI players face.
- HostDwarkesh Patel
- Episode Title: The data black hole at the center of AI
- Show: Dwarkesh Podcast
- Host: Dwarkesh Patel
- Length: 12 minutes
- Topics Covered: What is really driving AI progress? Comparing human vs AI sample efficiency. Does sample efficiency matter?
Read the transcript here.
Thanks to Mercury for sponsoring this essay!
Mercury just released a new feature called Command, which gives me AI right in my banking platform. And since I use Mercury to run basically my entire business, Command has access to all the info it needs to get real work done. I can ask it to send invoices, or categorize expenses, or even transfer money… and Command just handles it. Learn more at mercury.com/command
Timestamps:
(00:00:00) – What is really driving AI progress?
(00:03:11) – Comparing human vs AI sample efficiency
(00:08:46) – Does sample efficiency matter?
This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.dwarkesh.com
- sample efficiency
- Dwarkesh Patel
- how quickly AI learns from data compared to humans
1 h 371 h 201 h 37AI researchers debate how close we are to recursive self-improvement
WithJohn Schulman, Beren Millidge, Charlie O’Neill
2 h 21Questions people ask
- What is the main topic of the podcast?
- The episode discusses what is really driving AI progress and compares human versus AI sample efficiency.
- How long is the episode?
- The episode is 12 minutes long.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Dwarkesh Patel.